PSP's CPU not 333Mhz, only 222 ....suxx

[Simon F]

That strikes me as a very strange figure.

It's not strange, it's latency. The throughput is 16cycles, just like you would expect, and you can always execute invidual DOT4/MADDs single cycle if you prefer.

I'm surprised they would quote latency rather than throughput - I don't anyone else does and you generally code to try to hide the latency.

Anyway, in summary it can basically execute a MADD per clock (since that would then correspond to a matrix-vector mul throughput of 16 cycles).

The thing has a very impressive feature set though, it's not just about how fast it is (and it's quite fast, especially once you go beyond simple stuff like madds).

I'll take your word for it. I was just trying to get an idea of how it compares to a 'standard' DX/OGL vertex shader architecture.

Simon, that would mean that 4x4 Matrix * 4x1 Vector does not push the VFPU to the max (there migth be dedicated dto product instructions that yield that figure): for 333 MHz operation the CPU was and still is specced at 2.6 GFLOPS.

Again, that just sounds odd. A matrix-vector mul can be done as 4 DPs so if you can do a DP in under 4 clocks (i.e. to beat the the 16 clocks quoted by Fafalada) why would you use a suboptimal approach? It'd be madness.

Secondly, mat-vec muls are "bread and butter" operations in a graphics pipeline so you'd want to be damned sure they ran as fast as possible.

 

[cybamerc]

Shifty Geezer:

> Sony said the processor speed was scalable.

Yes. Scalable up to 333 MHz which it isn't the moment. Who cares about potential? For all intents and purposes this is a 222 MHz system. Yet Sony happily advertises the 333 MHz number.

> As for early adopters being at a disadvantage, the report suggests it's a
> matter of battery conservation

Yes, and what are those people going to do if Sony finally decides to unlock the full potential of the system. Battery life isn't good to begin with. At 333 MHz you'll likely have 1-2 hours of battery life. That isn't the number Sony is advertising.

So early adopters will either have to put up with awful battery life or pick up a new battery. If that isn't a disadvantageous situation I don't know what is.

Sony is being very underhanded.

 

[one]

Yet Sony happily advertises the 333 MHz number.

Source?
In the official spec sheet it claims 1-333Mhz. So technically you can't complain even if it's locked to 1 Mhz.

 

[cybamerc]

one:

> In the official spec sheet it claims 1-333Mhz. So technically you can't
> complain even if it's locked to 1 Mhz.

But you sure as hell can complain that it doesn't run at 333 MHz. And you sure as hell can complain that the battery life figures that Sony released aren't representative of what you'll once/if Sony unlocks the full potential of the system.

 

[Fafalada]

Simon, I agree that stating only latency is odd, but then these slides aren't without other errors, or misleading text... namely...

Again, that just sounds odd. A matrix-vector mul can be done as 4 DPs

I only just realized (damn lack of sleep and poor reading comprehension) what the real misunderstanding here is. The text on that slide is horribly misleading - the sample instruction in fact computes Matrix*Matrix, not Matrix*Vector.

 

[PC-Engine]

It'd be madness.

I guess you're not familiar with SONY MATH...

 

[Simon F]

I only just realized (damn lack of sleep and poor reading comprehension) what the real misunderstanding here is. The text on that slide is horribly misleading - the sample instruction in fact computes Matrix*Matrix, not Matrix*Vector.

Ahhh - now it makes sence.

 

[Qroach]

None of this crap matters, do you think people are going to make thier decision based on SONY not being honest about a performance number? Or that the battery life isn't going to be as much as advertised? All that matters is if it has good FUN games and lots of them. That's all the public wants and it's why the PSP is going to sell out very quickly (and end up out selling the DS). The matter of battery life is really an afterthought to the consumers.

 

[function]

Gotta agree with Qroach here.

99% of PSP owners or potential PSP owners will never be aware of this issue. As long as it has games they want, it'll be an attractive machine. Presumeably, by the time the full 333mHz is open to developers, new potential customers will be able to buy newer machines and the problem of reduced battery life will belong to the early adopters - most of whom will simply deal with it as long as they've enjoyed owning the device.

 

[Panajev2001a]

Fafalada:

http://common.ziffdavisinternet.com/util_get_image/7/0,1311,sz=1&i=78745,00.jpg

Two Sony presentations, two different e-DRAM bus widths for the GPU: what is the official/real value ?

Is it 256 bits or is it 512 bits ?

 

[Lazy8s]

Sony deems the battery life for a 333-MHz PSP at 90-nm to be unacceptable, yet the many millions of early adopters will be hit with that when the CPU is unlocked in later titles for future PSP hardware releases at 65-nm and below.

Fafalada:

The thing has a very impressive feature set

Seems like limited fixed functionality. A vertex shader at the level of just MS 1.0 would probably serve better than its support for morphing, skinning, etc.

 

[Qroach]

They aren't going to "unlock" any thing. I don't knowwhat give you that idea, but it's running at 222 mhz for a reason.

 

[Lazy8s]

The speculation is that they'll take off the cap of 222-MHz maximum performance on the programming interface in the future when chip shrinks and/or stronger batteries are available.

 

[Qroach]

Give me five good reasons why sony or any company would ever do that... sony may be able to do it, but they won't...

 

[Acert93]

Give me five good reasons why sony or any company would ever do that... sony may be able to do it, but they won't...

1. GB Next

2. Process shrink to 65nm allows new models to be more effecient.

3. Longer lasting batteries means new models do not take a hit on battery time at the higher frequency.

4. If developers hit the "wall" in PSP development, a50% increase in power could re-jumpstart development interestest.

5. Developers want to push the PSP to make better games.

6. GB Next

I am not sure if Sony WOULD do this, but in 2-3 years if they are facing some stiff competition and want to give the PSP a booste I could see this happening. Only new games would use the booste anyhow, so First gen models could still play games with the normal battery life.

It would just simple be explained that the "new games" suck the battery quicker because they are so much better. And that would be true... they are better because they run overclocked

Again, not sure if Sony would do this, but I think there is room to believe they could at least entertain the idea.

The problem I see is that battery use is not always linear. Going from 222MHz to 333MHz may be a 50% increase in processing power--but it may be a 100% or more increase in battery drain. So an intense game at 222MHz may last 5hrs, an intense game at 333MHz may only last 2-2.5hrs.

That right there is why I think it is currently locked at 222MHz. But in a couple years everyone will forgot all about this and when new games suck the battery quicker it will be just explained as "The new games are really intense". That is, if they decide to unlock the cap.

 

[Lazy8s]

They'll think they have an excuse to do it when new PSP technology which can support >222-MHz operation acceptably becomes the standard.

 

[hey69]

It was confirmed in last week's Edge magazine, with Ridge Racers' associate producer Hideo Teramoto, that Ridge Racers was running underclocked and that future Namco titles will be using the maximum 333 MHz speed.

somebody has Edge magazine around?

 

[London Geezer]

I don't see why people make such a big fuss about this.

Some games, the ones that require it, will run at the full 333MHz, some at less, according to their needs.

Namco said ages ago that their games (will or already) run at 333MHz.

If a developer feels he can get away with using less than that, he will do so in exchange for more battery life. If anything he can exchange battery life with disc spinning management.

 

[-tkf-]

Give me five good reasons why sony or any company would ever do that... sony may be able to do it, but they won't...

They just need one good reason, enough PSP's in the Wild and a general feeling that the battery is lasting long enough.

I could see GT4 using 333Mhz when it's released, battery time won't be as big an issue when the PSP is widespread, instead of blamming the machine the customers will blame the game.

 

[Fafalada]

Seems like limited fixed functionality. A vertex shader at the level of just MS 1.0 would probably serve better than its support for morphing, skinning, etc.

I was talking about the featureset of the VFPU, relative to other SIMD solutions in consumer space.

As for GPU, I agree that Vertexshader is preferable over fixed functions, but the latter obviously has a better performance/cost ratio, and IMO for portable hw the ease of use generally comes ahead of flexibility.

If you look at their desktop "predecessors", both DS and PSP follow the idea of simplification while maintaining similar performance characteristics.